MadGraph5_aMC@NLO

Matrix element of the process g g > g g h h

Asked by Shi-Ping He

Dear MadGraph developers,

I am testing the g g > g g h h process provided in V. Hirschia and O. Mattelaer's paper (arXiv:1507.00020). After changing kinematic configuration into that in this paper, I obtained the following results:
Phase space point:
     ---------------------
     E | px | py | pz | m
     0.50000000000000000E+03 0.00000000000000000E+00 0.00000000000000000E+00 0.50000000000000000E+03 0.00000000000000000E+00
     0.50000000000000000E+03 0.00000000000000000E+00 0.00000000000000000E+00 -0.50000000000000000E+03 0.00000000000000000E+00
     0.14855661132240300E+03 -0.20035064773965502E+02 0.36334271097681800E+02 -0.68720329531360505E+02 0.12499999999999963E+03
     0.32282497237901401E+03 -0.94143380761058395E+02 -0.27371372875378103E+03 0.69345377222217806E+02 0.12499999999999777E+03
     0.13811806564789200E+03 -0.95984870048668895E+02 -0.88577275525737207E+02 0.44917380160615302E+02 0.72773677908859577E-05
     0.39050035065069000E+03 0.21016331558369299E+03 0.32595673318183600E+03 -0.45542427851472603E+02 0.17460974977252652E-04
     Four-momentum conservation sum:
     0.99475983006414026E-12 -0.19895196601282805E-12 0.43343106881366111E-12 0.00000000000000000E+00
    ---------------------
 -------------------------------------------------
 Matrix element = 3.8670886931918955E-013 GeV^ -4
 -------------------------------------------------
This is 3.37 times smaller than the result 1.3033633142042775e-12 GeV^-4 in V. Hirschia and O. Mattelaer's paper. Maybe there are some differences on the parameters and settings, but such discrepancy is too big to explain.

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

The installed MadGraph version is "MG5_aMC_v3_5_3". Here, I copied some details for identification:

External Params
  ---------------------------------

 MU_R = 91.188000000000002
 aEWM1 = 132.50700000000001
 mdl_Gf = 1.1663900000000000E-005
 aS = 0.11799999999999999
 mdl_ymb = 4.7000000000000002
 mdl_ymt = 173.00000000000000
 mdl_ymtau = 1.7769999999999999
 mdl_MT = 173.00000000000000
 mdl_MB = 4.7000000000000002
 mdl_MZ = 91.188000000000002
 mdl_MH = 125.00000000000000
 mdl_MTA = 1.7769999999999999
 mdl_WT = 1.4915000000000000
 mdl_WZ = 2.4414039999999999
 mdl_WW = 2.0476000000000001
 mdl_WH = 6.3823389999999999E-003

===============================================================
  > MLReductionLib = 7|6|1
  > CTModeRun = -1
  > MLStabThres = 1.0000000000000000E-003
  > NRotations_DP = 0
  > NRotations_QP = 0
  > CTStabThres = 1.0000000000000000E-002
  > CTLoopLibrary = 2
  > CTModeInit = 1
  > CheckCycle = 3
  > MaxAttempts = 10
  > UseLoopFilter = F
  > HelicityFilterLevel = 2
  > ImprovePSPoint = 2
  > DoubleCheckHelicityFilter = T
  > LoopInitStartOver = F
  > HelInitStartOver = F
  > ZeroThres = 1.0000000000000001E-009
  > OSThres = 1.0000000000000000E-008
  > WriteOutFilters = T
  > UseQPIntegrandForNinja = T
  > UseQPIntegrandForCutTools = T
  > IREGIMODE = 2
  > IREGIRECY = T
  > COLLIERMode = 1
  > COLLIERRequiredAccuracy = 1.0000000000000000E-008
  > COLLIERCanOutput = F
  > COLLIERComputeUVpoles = F
  > COLLIERComputeIRpoles = F
  > COLLIERGlobalCache = -1
  > COLLIERUseCacheForPoles = F
  > COLLIERUseInternalStabilityTest = T
 ===============================================================
It also reminds that " Onshellness of the momentum of particle 3/4/5/6 of mass is not accurate enough.

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

Best,
Shi-Ping He

Question information

Language:
English Edit question
Status:
Solved
For:
MadGraph5_aMC@NLO Edit question
Assignee:
Valentin Hirschi Edit question
Solved by:
Shi-Ping He
Solved:
Last query:
Last reply:
Revision history for this message
Olivier Mattelaer (olivier-mattelaer) said : 		#1

Hi,

I have tested this today with the LTS and 3.5.3.

When I run via ./check within the P directory I do get the following result:
 Matrix element finite = 1.3033633178489090E-012 GeV^ -4 (for LTS)
 Matrix element finite = 1.3033633178267480E-012 GeV^ -4 (for 3.5.3)
 Matrix element finite = 1.3033633142042775e-12 (from the paper --in quadruple precision--)

Now if I use the "launch" command from MG5aMC
3.5.3 returns a slightly different number:
1.2391506868028831e-12

This seems related the madloop entry "DoubleCheckHelicityFilter" which is set to False automatically when using launch.
I will let Valentin comment on that quite technical point.

But at least I do not reproduce your missmatch and therefore can not comment on it.
(PS: I have check the stability of the result comparing collier/ninja/Cuttools for the reduction tools and none of them seems problematic.

Cheers,

Olivier

Revision history for this message
Shi-Ping He (shi-pinghe) said : 		#2

Dear Olivier,

Thanks very much for your reply.

Could you check the result for the automatically generated kinematic configuration? If our results agree, maybe there are some mistakes when I modified the "check_sa.f".

For the model "loop_qcd_qed_sm", here are the automatic results in my computer:

|================================================================================================
|| Results for process gg > gghh (Loop-induced)
|================================================================================================
|| Phase-Space point specification (E,px,py,pz)
|
| 5.0000000000000000e+02 0.0000000000000000e+00 0.0000000000000000e+00 5.0000000000000000e+02
| 5.0000000000000000e+02 0.0000000000000000e+00 0.0000000000000000e+00 -5.0000000000000000e+02
| 8.2699870436642712e+01 -2.0640279037102381e+01 3.7431847739289353e+01 -7.0796216186535759e+01
| 3.0663344631872172e+02 -9.6987240636697933e+01 -2.8198200512458772e+02 7.1440145162875041e+01
| 1.8939780888763451e+02 -9.8884357175572646e+01 -9.1252995876172164e+01 4.6274233230103313e+01
| 4.2126887435700098e+02 2.1651187684937301e+02 3.3580315326147058e+02 -4.6918162206442616e+01
|
| Stable kinematic configuration (SPS).
| Estimated relative accuracy = 1.2e-11
|
|| Loop amplitude squared, must be finite:
| Finite = 4.8803673752777677e-11
|(| Pole residues, indicated only for checking purposes: )
|( Single pole = 9.6244043097511608e-24 )
|( Double pole = 2.6170076754473879e-35 )
|
|
|| All virtual contributions are of split orders *(QCD=8)
|================================================================================================

Best,
Shi-Ping He

Revision history for this message
Olivier Mattelaer (olivier-mattelaer) said : 		#3

Hi,

In the loop_sm/version 3.5.3, I do get:
| 5.0000000000000000e+02 0.0000000000000000e+00 0.0000000000000000e+00 5.0000000000000000e+02
| 5.0000000000000000e+02 0.0000000000000000e+00 0.0000000000000000e+00 -5.0000000000000000e+02
| 8.2699870436642712e+01 -2.0640279037102381e+01 3.7431847739289353e+01 -7.0796216186535759e+01
| 3.0663344631872172e+02 -9.6987240636697933e+01 -2.8198200512458772e+02 7.1440145162875041e+01
| 1.8939780888763451e+02 -9.8884357175572646e+01 -9.1252995876172164e+01 4.6274233230103313e+01
| 4.2126887435700098e+02 2.1651187684937301e+02 3.3580315326147058e+02 -4.6918162206442616e+01
|
| Stable kinematic configuration (SPS).
| Estimated relative accuracy = 1.8e-08
|
|| Loop amplitude squared, must be finite:
| Finite = 4.5743546589142221e-11
|(| Pole residues, indicated only for checking purposes: )
|( Single pole = 0.0000000000000000e+00 )
|( Double pole = 0.0000000000000000e+00 )

(both with check and launch (I did also get that result with the launch command in LTS)

In the loop_qcd_qed_sm/LTS
|| Loop amplitude squared, must be finite:
| Finite = 4.8833261713423888e-11
 (confirmed with the launch command in 3.5.3, did not run it via the check command)

Now can you confirm that you set the order of the particle correctly.
In the paper, the momenta are given in the "g g > h h g g" order (and in your first post the momenta are also given in that order.
But to reproduce your number above, I had to use the "g g > g g h h" syntax.
This could explain your large difference.

Cheers,

Olivier

PS: I discussed the issue between check/launch result and it seems that a threshold effect in one optimization which is not secure enough. I still need to investigate this a bit more to understand when this was introduced (since I do not observe that in our LTS version) and then check various method to make that optimization more secure/efficient.

> On 11 Jan 2024, at 10:05, Shi-Ping He <email address hidden> wrote:
>
> Question #708917 on MadGraph5_aMC@NLO changed:
> https://answers.launchpad.net/mg5amcnlo/+question/708917
>
> Status: Answered => Open
>
> Shi-Ping He is still having a problem:
> Dear Olivier,
>
> Thanks very much for your reply.
>
> Could you check the result for the automatically generated kinematic configuration? If our results agree, maybe there are some mistakes when I modified the "check_sa.f".
>
> For the model "loop_qcd_qed_sm", here are the automatic results in my computer:
>
> |================================================================================================
> || Results for process gg > gghh (Loop-induced)
> |================================================================================================
> || Phase-Space point specification (E,px,py,pz)
> |
> | 5.0000000000000000e+02 0.0000000000000000e+00 0.0000000000000000e+00 5.0000000000000000e+02
> | 5.0000000000000000e+02 0.0000000000000000e+00 0.0000000000000000e+00 -5.0000000000000000e+02
> | 8.2699870436642712e+01 -2.0640279037102381e+01 3.7431847739289353e+01 -7.0796216186535759e+01
> | 3.0663344631872172e+02 -9.6987240636697933e+01 -2.8198200512458772e+02 7.1440145162875041e+01
> | 1.8939780888763451e+02 -9.8884357175572646e+01 -9.1252995876172164e+01 4.6274233230103313e+01
> | 4.2126887435700098e+02 2.1651187684937301e+02 3.3580315326147058e+02 -4.6918162206442616e+01
> |
> | Stable kinematic configuration (SPS).
> | Estimated relative accuracy = 1.2e-11
> |
> || Loop amplitude squared, must be finite:
> | Finite = 4.8803673752777677e-11
> |(| Pole residues, indicated only for checking purposes: )
> |( Single pole = 9.6244043097511608e-24 )
> |( Double pole = 2.6170076754473879e-35 )
> |
> |
> || All virtual contributions are of split orders *(QCD=8)
> |================================================================================================
>
>
> Best,
> Shi-Ping He
>
> --
> You received this question notification because you are an answer
> contact for MadGraph5_aMC@NLO.

Revision history for this message
Shi-Ping He (shi-pinghe) said : 		#4

Dear Olivier,

Thanks very much for your kind and patient help.
It is indeed caused by the particle order. After changing "g g > g g h h" to "g g > h h g g", I can reproduce your results.

Best,
Shi-Ping He

Revision history for this message
Olivier Mattelaer (olivier-mattelaer) said : 		#5

Just updating on the issue detected in 3.5.3. This was finally a compiler specific issue related to the rambo phase-space integrator used for the initialisation. This has been fixed here: https://github.com/mg5amcnlo/mg5amcnlo/commit/c28416c08b6a12f6fae38238dcab3ade7024290a

Thanks,

Olivier